Skip to content

Latest commit

 

History

151 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Thinking Map

CI License: MIT

An AI thinking partner that helps you deconstruct a vague idea and turn your thinking into a visual map and an actionable plan.

You arrive with something you cannot yet describe — "I want to build an educational game for kids, but I don't know what it should be" — and instead of answering, the partner asks the two or three questions that would change what you should build. Every answer becomes a node on the map. You leave with what you know, what you don't, the strongest directions, and the smallest thing worth building first.

The central principle: don't just give me an answer — help me understand the problem well enough to find a better answer.

The map moves through five phases: 01 Idea → 02 Map → 03 Research → 04 Explore → 05 Next steps.

Run it locally

npm run setup   # install dependencies, provision PostgreSQL, push the schema, seed
npm run dev     # http://localhost:3000

Nothing has to be installed or hosted first. With no DATABASE_URL set, setup starts a local PostgreSQL server of its own and writes the connection string to .env.local. Set DATABASE_URL yourself and that database is used instead, untouched. See DATABASE.md.

No API key is needed. The app never calls a model itself — the agent is the one you already have, and it brings its own credentials. See AGENT_CONTRACT.md for the three ways an agent can attach.

Develop this project with codeyam-editor

This project is built with codeyam-editor — code and runnable data scenarios are authored side by side against a live preview.

# Clone the repo
git clone https://github.com/codeyam-ai/thinking-map && cd thinking-map

# Install codeyam-editor
npm install -g @codeyam-editor/codeyam-editor@latest

# Launch the editor (split-screen terminal + live preview)
codeyam-editor start

Every screen has runnable scenarios carrying their own seed data, so any state can be viewed without touching real data:

codeyam-editor editor scenarios       # list every registered scenario
codeyam-editor editor refresh-tests   # run the test suite
npx tsc --noEmit                      # type-check

What it looks like

Each of these is a registered scenario — a real state of the app, captured.

The way in

A single yellow card on a black screen asking "What are you trying to figure out?"

One card, one free-text box. No structured fields. You can also attach a brief — browse for a .pdf / .docx / .md / .txt / .html, or point at a page — and with one attached the sentence becomes optional.

The map mid-round

A full-height dark board: one idea card on the left, four coloured branches fanning right into rows of question and finding cards

Your idea sits on the left. Each coloured branch is a theme the partner pulled out of it, and the cards hanging off it are what it wants to know. A question and the node it becomes are the same card — you answer inside the map rather than reading the question in one place and answering it in another. The bar along the top always says what is still waiting on you.

One map, both hands on it

The board mid-exchange, with cards contributed by the agent and by the person on the same branches

The partner writes to the map and so do you, and the map does not distinguish between the two by putting them in separate places. An answered question and a node the agent added sit on the same branch, because what matters is where a thought belongs, not who had it.

Arriving with a brief

The board seeded from an attached brief, with the source document standing behind the first card

A twenty-page spec is stored whole as the map's source. The partner reads it the way anyone reads a long document — the outline first, then the passages that matter — so a long brief cannot quietly fill the context that ought to be spent thinking about it.

The plan you leave with

A build sequence where one slice is marked "proves nothing yet", above a five-step track

A plan is a build sequence, not a to-do list. Each increment names the assumption or open question that building it would settle — and one that settles nothing is marked proves nothing yet rather than sitting in the sequence looking like progress.

What to do next

The summary screen: next five steps as a left-to-right track, with the activity log below

The end of the loop: what you know, what you don't, the directions worth taking, and five concrete steps in order — plus the activity log of everything that happened to the map, from both sides.

Notes for anyone picking this up

  • package.json runs next dev --webpack. Next 16 defaults to Turbopack, and Turbopack's dev output does not hydrate through the codeyam preview proxy. This is deliberate, not a leftover.
  • A plain npm run dev does not serve /isolated-components/*, the fixture pages scenario captures render from — they would fill your dev session with convincing fake maps. Use CODEYAM_APP_PORT=1 npm run dev if you want them.
  • Before adding a feature that touches auth, file uploads, email, or another external service, read FEATURE_PATTERNS.md.

Contributing

Issues and pull requests are welcome. CONTRIBUTING.md covers getting set up, the checks a change has to pass, and the few conventions here that look like mistakes and are not. Everyone taking part is expected to uphold the Code of Conduct; to report a vulnerability, see SECURITY.md rather than opening an issue.

License

MIT © 2026 CodeYam.

Scenario gallery

States captured as runnable scenarios with codeyam-editor:

A board with pictures, a drawn shape and a shortlist

A board with pictures, a drawn shape and a shortlist

A plan with a gap - one slice proves nothing

A plan with a gap - one slice proves nothing

A brief, and nobody has picked it up yet

A brief, and nobody has picked it up yet

Complete - what to do next

Complete - what to do next

All eight, the list expanded

All eight, the list expanded

Weighing the alternatives against each other

Weighing the alternatives against each other

An older question, three rounds up, still open

An older question, three rounds up, still open

Brief attached, nothing cited yet

Brief attached, nothing cited yet

About

An AI thinking partner that deconstructs a vague idea, explores the problem space, and turns your thinking into a visual map and an actionable plan.

Topics

Resources

Code of conduct

Contributing

Security policy

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages